Crate eidos_kernel 

eidos-kernel — the pure-logic brain engine.

This crate is L1 of the Eidos layer model. It owns the graph schema, the deterministic retrieval engine (BM25F with calibrated confidence bands), the golden-set eval harness, and the ranking policy (floor, scope, canonical dominance). It has no IO — no tokio, no rmcp, no clap, no std::fs. Every function is a pure transformation over Graph, fully unit- testable with zero files on disk.

The grounding law lives here: a source proposes nodes and edges; the kernel owns and verifies every receipt — each file:line is checked against the real file before a hit is returned. Sources compute nodes; the kernel never fabricates provenance.

Modules

module	owns
schema	Node, Edge, Graph, Kind, Span, relation vocabulary
retrieval	ground, ground_scoped, BM25F scoring, confidence bands
eval	the golden-set eval harness (GoldenSet, evaluate)
relational	callers, impact, kind-intent, rerank-by-kind
workflow	deterministic task brief decomposition and validation

Modules

calibration

The calibration surface — the single documented home for the constants that shape ranking and confidence. Moving any value here (or an inline boost listed in the REGISTRY below) moves the frozen eval baselines, so every change is a reviewed diff against evals/baseline.json plus, in the ideal, a re-freeze justification.

eval

Retrieval eval — score ground against a golden set so graph changes are MEASURED, not asserted. Mirrors the discipline of the Python test_retrieval_eval, but runs over whatever graph it’s handed (a hermetic fixture in tests; the real vault from the CLI).

graph_index

Indexed read view over a Graph.

relational

Relational (call-graph / impact) retrieval over the code graph.

retrieval

Retrieval — lexical scoring + CALIBRATED confidence.

schema

Graph schema — nodes, edges, kinds. Defines the Aun graph-artifact contract.

trigram

Trigram inverted index — the hippocampus (fast pattern separation).

workflow

Deterministic task decomposition primitives.